CARVE, EpicKitchen-AR
收藏资源简介:
CARVE数据集由迪肯大学应用人工智能研究所创建,包含10,000个通过2D物理模拟器生成的合成视频,涵盖了250,000多个因果事件对。该数据集通过反事实合成方法生成触发-目标事件标签,旨在解决视频事件中的因果推理问题。EpicKitchen-AR数据集则基于EpicKitchen-100数据集中的真实视频,通过反事实合成方法增加了因果事件标签,扩展了CARVE任务在真实世界视频中的应用。这两个数据集为视频因果推理研究提供了丰富的实验数据,适用于视频监控、根因分析和电影内容管理等领域。
The CARVE dataset was developed by the Institute of Applied Artificial Intelligence at Deakin University. It contains 10,000 synthetic videos generated using a 2D physics simulator, covering more than 250,000 causal event pairs. The dataset generates trigger-target event labels via counterfactual synthesis methods, with the goal of solving causal reasoning tasks in video events. The EpicKitchen-AR dataset, based on real videos from the EpicKitchen-100 dataset, adds causal event labels through counterfactual synthesis, extending the application of the CARVE task to real-world video scenarios. Collectively, these two datasets provide abundant experimental data for video causal reasoning research, and are applicable to fields such as video surveillance, root cause analysis, and film content management.




